On-premise assistants can give factory workers conversational access to machine documentation, but models capable of the task rarely fit shop-floor hardware. We show that after structural compression and retrieval-grounded adaptation, model size is no longer a reliable predictor of adapted answer quality: general capability falls almost linearly with parameter count, while judged retrieval-augmented answer quality does not. We therefore treat deployment as a post-adaptation selection problem, committing one sub-network per device on judged answer quality and measured on-device throughput under a configurable general-capability floor and memory budget; rules that optimize size, speed, or quality alone each give up capability or throughput. A weight-shared supernetwork trained with sandwich-style in-place distillation keeps this selection inexpensive. In a manufacturing-manual case study, extraction costs 13.7 percent of the unpruned model's judged quality and retrieval-grounded distillation returns it to within 4.6 percent, recovering two thirds of the loss, and the same assistant runs across three heterogeneous edge tiers at 1.3 to 5 watts standby.
Deploying 3D point cloud models on edge hardware such as the NVIDIA Jetson Orin Nano is severely constrained by compute and memory budgets. Existing compression methods require access to the model's original source code, rendering them inapplicable to the Open Neural Network Exchange (ONNX) binaries commonly distributed by vendors and model repositories. We present \textbf{H3DNAS}, a hardware-aware model compression framework that operates directly on ONNX computational graphs without requiring original source code, architecture class definition, or gradient access during search. H3DNAS makes three contributions: (1) a \textbf{Channel Dependency Graph (CDG)} that classifies ONNX operators into four constraint classes and formally establishes that the free parameter fraction $ρ_f$ is topological invariant, a provable compression ceiling computable in $\mathcal{O}(|V|+|E|)$; (2) a \textbf{Two-Stage Hierarchical Search} that prunes candidate architectures by $L_1$-importance channel selection, ranks them by output fidelity as a zero-shot label-free proxy, and applies GhostConv structural mutation to Pareto-optimal candidates; and (3) the \textbf{first source-code-free compression pipeline for 3D point cloud models}, operating entirely via ONNX graph surgery with no original architecture definition required. On ModelNet40, H3DNAS reduces the number of parameters in PointNet, PointNet++, and PointMLP by $65.5\%$, $43.2\%$, and $49.1\%$, respectively, while achieving $1.99\times$, $1.29\times$, and $1.67\times$ inference speedups with negligible loss in accuracy. The source code is publicly available\footnote{https://github.com/ClarityLab-Org/h3dnas}.
Ao Qu, Panagiotis Michelakis, Linyuan Han +8cs.AI cs.RO
This paper presents FAIRY, a full-stack smart-agriculture agent system developed for and deployed to an operating soybean research farm at Harbin Institute of Technology's smart-agriculture site. We develop FAIRY to execute and evaluate agentic agronomic operations on full-season spatiotemporal workflows that span ridge preparation, planting, irrigation, fertilization, pest and disease treatment, harvest, grain handling, drying, and storage. FAIRY integrates APIs and infrastructure across production-grade machinery, fixed soil and canopy sensors, multispectral and thermal drones, satellite vegetation products, a weather station, calibrated crop-process models, agronomic records, and multi-season yield histories. The system is built around the novel "everything is an event" execution paradigm, which represents spatiotemporal world evolution, remote sensing and UAV observations, sensor readings, crop-growth transitions, machinery actions, and management interventions as state-changing events in a shared farm process engine. On top of this event-driven world model, FAIRY implements a complete agentic stack: a knowledge library of atomic agronomic skills; multi-agent controller and orchestration backends; frontier- and edge-model execution; full-path trace logging; and deployment profiling on local nodes. We use FAIRY to evaluate nine state-of-the-art agent controllers across one hundred full-season soybean scenarios that preserve the operational coupling between spatial observations in a 64-ridge field, temporal decision sequences, agronomic constraints, delayed effects, and final yield. We develop an evaluation suite that combines agentic success, full-path spatiotemporal correctness, token cost, and edge-device runtime.
Pothole detection and its severity measurement is still an important challenges in urban infrastructure management, where late maintenance directly contributes to vehicle damage, road accidents, and escalating repair costs. Existing automated approaches depend on 2D RGB images and cannot measure physical depth of potholes. In this paper, we present a depthaware pothole detection framework and then compare five architectures: YOLOv8n, YOLOv8nSeg, YOLOv9t, RTDETRL, and RTDETRX for RGB-D sensor fusion-based detection and automated depth measurement. A custom offline augmentation pipeline is used here to simulate adverse road monitoring conditions. All models are trained on the PothRGBD dataset with an 80% training and 20% validation split and evaluated using Precision, Recall, mAP@50, and mAP@50_95. Before measuring the depth data, all depth maps are corrected for camera tilt using RANSAC ground-plane orthorectification and all zero-valued sensor pixels are cast to NaN before any statistic is computed. YOLOv8nSeg achieves the highest mAP@50 of 0.9556 and mAP@50_95 of 0.6758 with the most accurate depth estimate of 2.96 cm with the pixel-precise Dseg algorithm. YOLOv8n achieves the fastest inference at 3.6ms. RTDETRX achieves the highest detection confidence at 92.70%. An important finding is that even after full RANSAC orthorectification, bounding box models overestimate pothole depth by 0.16 to 0.21 cm compared to pixel precise segmentation masks. This confirms that the pavement inclusion bias is structural rather than a calibration artifact.
Automated tea leaf disease classification supports precision agriculture, yet deploying accurate models on edge devices remains challenging under tight compute budgets. Self-supervised vision foundation models such as DINOv2 provide strong features but are too large for field deployment, while lightweight models trained from scratch on small agricultural datasets often underfit. We study cross-architecture knowledge distillation (KD) from a fine-tuned DINOv2 teacher (Vision Transformer) to a compact bidirectional Visual State Space Model (LVSSM) student, an underexplored direction because the architectures use fundamentally different token-mixing mechanisms. We identify and fix two training-stability problems that prevent the from-scratch SSM student from learning on limited data: a single large patch-embedding convolution and a fusion layer that severs the residual path. With a progressive convolutional stem and gated bidirectional selective-scan block, the 4.45M-parameter student trains stably. Across three seeds, temperature-scaled logit distillation raises test accuracy from 92.32+/-2.14% to 95.41+/-1.17% (best single run: 96.20%; macro-F1: 94.45%), a +3.09 percentage-point mean gain. The student uses 5.0 times fewer parameters than the 22M-parameter teacher while retaining 98.3% of its accuracy. Ablations show that intermediate feature-alignment losses reduce accuracy, making simple logit-level KD the strongest configuration. A fair from-scratch comparison shows the gain is specific to students that start below the teacher. We report per-class metrics, confusion matrices, bootstrap confidence intervals, and FLOPs/latency measurements, and discuss limitations including the single-dataset scope and simplified non-official SSM implementation.
Extreme compression of deep neural networks, up to full binarization, dramatically reduces memory footprint and arithmetic complexity, facilitating deployment on constrained edge hardware with field-programmable gate arrays (FPGAs) and microcontrollers. Although combining binarization with pruning promises additional efficiency gains, existing pruning strategies are ill-suited to binarized representations and rarely translate into meaningful hardware savings. We introduce a PyTorch-based, research-oriented framework that incorporates freezing and pruning mechanisms for designing and optimizing binarized neural networks. The framework enables rapid and reproducible evaluation of state-of-the-art approaches and the fast prototyping of new ones. Leveraging this framework, we propose a novel pruning method that accounts for the relative importance of learned parameters across abstraction levels. Such a global weighting mechanism consistently achieves a superior trade-off between model accuracy and pruning rate, achieving a 70% pruning rate on VGG11 with constant accuracy, while state-of-the-art results reach only 41% in the binarized setting.
Large-kernel Convolutional Neural Networks (CNNs) deliver remarkable performance in vision tasks by significantly expanding receptive fields, yet their quadratic parameter growth critically impedes storage-efficient edge deployment. While existing efficient architectures adopt parameter-efficient depthwise separable convolution backbones that leverage techniques like low-rank approximation and weight sharing to compress depthwise convolutions, we identify a critical oversight: pointwise convolutions dominate parameter volume (>87% in models like RepLKNet-31B) and constitute the primary deployment bottleneck on resource-constrained edge devices. This results in prohibitive storage costs and severe memory-loading constraints on resource-limited devices (e.g., smartphones with 4-12 GB Random Access Memory (RAM)). To overcome this, we propose Channel Group-Shared (CGS) low-rank approximation, a novel Singular Value Decomposition (SVD)-based parameter-sharing strategy. CGS constructs a structured low-rank paradigm isomorphic to SVD decomposition, comprising shared (high-parameter-cost) down/up-projection matrices across channel groups within a layer and channel-group-specific (low-parameter-cost) scalable diagonal matrices. This group-sharing design achieves significant parameter reduction. Extensive experiments demonstrate that large-kernel CNNs (RepLKNet, ConvNeXt, SLaK) enhanced with CGS strike an empirically favorable balance between competitive performance and substantially reduced storage costs. Crucially, by alleviating storage constraints, reducing memory bandwidth pressure during loading, and minimizing model loading latency, CGS enables the feasible deployment of pre-trained large-kernel CNN models on edge devices, thereby bridging the gap between high-performance vision models and practical edge deployment.
Transformer-based object trackers are renowned for their strong performance, yet dense token processing often leads to prohibitive computational cost, limiting real-time deployment on edge devices. While recent works explore token pruning to reduce computation, they often stop short of an end-to-end sparse pipeline, as early-layer token scores can be noisy without a motion prior, and many trackers ultimately fall back to dense reshaping to feed the dense prediction head that partially negates the savings. We introduce Motion-aware Sparse Tracker (MaST), a sparse tracking framework that makes sparsity effective from tokens to boxes. First, MaST injects a lightweight motion prior to refine cross-attention-based importance scores, enabling earlier and more stable token reduction in the search region. Second, we introduce a natively sparse prediction head that operates directly on the retained unstructured tokens with a score-first, regress-once design, eliminating dense padding/reshaping and reducing redundant computation. Extensive experiments on multiple benchmarks demonstrate that MaST establishes new state of the art among lightweight trackers, where MaST-tiny attains 63.8 AUC on LaSOT and 80.1 SUC on TrackingNet, surpassing the prior best AsymTrack-S by +1.0 AUC and +2.2 SUC while running at 152 FPS on Jetson Nano, nearly twice as fast as AsymTrack-S at 88 FPS. Code is available at https://github.com/TsingWei/MaST.
William Howes, Farid Ahmed, Syed Bahauddin Alamcs.LG
Virtual sensing enables digital twins and safety-critical systems to reconstruct and forecast spatial-temporal physics in real time. However, conventional computational and data-driven methods often face challenges in generalization, latency, and energy efficiency for edge deployment. Neural operators offer a promising alternative but remain reliant on power-intensive hardware. Spiking neurons and neuromorphic computing can improve efficiency, yet surrogate-gradient training and multi-step spiking introduce convergence and latency challenges. We propose the Sparse-Activation-ReLU (SAR) layer, a single-step alternative that promotes activation sparsity without surrogate-gradient training while remaining compatible with event-based computing. Within a trunk-based NOMAD architecture, SAR achieves over a fivefold improvement in the combined Latency-Error-Energy (LEE) metric compared with Variable Spiking Neuron (VSN) and Leaky Integrate-and-Fire (LIF) implementations. We further analyze spiking entropy and feature usage and introduce synthetic knowledge distillation, reducing the LEE score by more than twofold. Finally, we improve VSN through a ReLU-based spiking loss and graph-neighbor thresholding. On the Heat Exchanger dataset, these approaches reduce L2 error by more than twofold and nearly sevenfold, respectively, while reducing spiking and spatial aggregation. Overall, the work presented is a step towards energy-efficient virtual sensing by providing an alternative framework that can be positioned towards neuromorphic or other edge device integration that can be a gold standard to compare latency, energy, and error performance for future efficient designs that are sparsity or brain-inspired spiking based.
Ranjan Sapkota, William Bu, Chen Chen +2cs.CV cs.AI
Accurate identification of early-stage apple fruitlet anatomical structures, including the calyx, fruitlet body, and peduncle, is essential for robotic thinning, crop-load management, and other precision orchard operations. This study presents a lightweight multimodal vision-language framework that adapts TinyCLIP for fine-grained fruitlet anatomy classification in complex orchard environments. A dataset of 600 high-resolution RGB images collected from Scilate and Scifresh apple orchards was converted into 224 x 224 image patches and annotated for three anatomical classes. Domain-specific language prompts, such as ``a photo of a class,'' were used to guide multimodal alignment between orchard imagery and horticultural structures. A sliding-window inference strategy with a stride of 112 pixels aggregates patch-level predictions into spatial heatmaps, enabling interpretable whole-image localization of fruitlet components relevant to robotic thinning. Patch-level evaluation on an NVIDIA T4 GPU achieved F1-scores of 0.95 for calyx, 0.98 for fruitlet, and 0.85 for peduncle, with a macro-F1 score of 0.93. Deployment-oriented optimization using ONNX and TensorRT enabled efficient inference on NVIDIA Jetson hardware, preserved accuracy under INT8 quantization, and supported model sizes of approximately 127-137 MB with millisecond-level patch inference. These results demonstrate that lightweight vision-language models can provide interpretable and edge-deployable perception for automated fruitlet analysis and future robotic thinning systems. The source code and implementation details are publicly available at https://github.com/WilliamBu1/A-Lightweight-Vision-Language-Model-for-Early-Stage-Fruitlet-Classification-in-Apple-Orchards.
Yalda Taheri, Mohammad Hassan Heydari, Erfan Naaman +1cs.AI cs.CL
Function calling represents the core capability of agentic large language models (LLMs). Existing research has focused on enhancing LLMs function-calling accuracy through fine-tuning, reinforcement learning (RL), and multi-agent frameworks, particularly for native function-calling LLMs. This work demonstrates that LLMs achieve superior accuracy in function calling in instruction-following contexts (i.e., standard user-assistant interactions) rather than a tool calling context. We introduce Instruction-Followed Function Calling (IFFC), a novel framework that decouples function-calling logic from the primary LLM and delegates it to a dedicated smaller model operating within the instruction-following paradigm. Our method consistently outperforms both native function calling (NFC) and prompt-based function calling (PFC) baselines, with particularly strong gains on reasoning-oriented LLMs. Furthermore, we demonstrate that IFFC maintains robust performance under aggressive quantization, enabling efficient on-device deployment without significant accuracy degradation. This work establishes a new paradigm for reliable, resource-efficient function calling in edge-computing scenarios.
Oral cancer is a leading cause of mortality in low-to-middle-income countries, where a shortage of specialists delays diagnosis. While point-of-care screening via smartphones offers a scalable solution, developing robust AI for resource-constrained settings poses significant challenges, including class imbalance in training data, variable data quality, and computational constraints on edge devices. In this paper, we present the optimisation of lightweight deep learning models for smartphone-based oral cancer screening. Using a diverse, multi-centre retrospective dataset of approximately 30,000 images acquired over a decade, we systematically evaluate state-of-the-art convolutional, transformer, and hybrid architectures. Through rigorous pipeline ablation, we demonstrate that directly optimising hybrid architectures for the edge strictly outperforms computationally heavy paradigms, such as large models or knowledge distillation. Furthermore, interpretability analysis and simulated noise-stress tests revealed that the system anchors on clinical features and remains robust to unstructured sensor noise, despite vulnerabilities to impulse bit errors. In the held-out test set, our optimised MobileViTv2 models achieved an average sensitivity of 83.2 $\pm$ 1.5% and an average specificity of 86.0 $\pm$ 0.8%, with the best model exhibiting 87.4% sensitivity, 86.5% specificity, and a critical negative predictive value of 97.2% with reference to specialist labels. These results confirm that with targeted architectural selection and streamlined optimisation, interpretable and robust lightweight AI models exhibit high potential for edge deployment to enable automated triage in primary care settings.
Rishiraj Sengupta, Sotiris Chatzimiltis, Mohammad Shojafar +1cs.CL cs.AI cs.NI
Real-world fault analysis in 5G and emerging 6G networks demands domain expertise to analyze free-text diagnostics, including root-cause explanations and recommended actions. LLMs have emerged as a promising approach to automating this, yet whether lightweight, edge-deployable models are capable of performing in-depth free-text diagnostics remains an open question. While existing benchmarks rely on restrictive MCQs with fixed answer keys, this paper evaluates 5G domain understanding and fault analysis in a free-text generation format. Transitioning to this paradigm requires evaluating lightweight, edge-deployable AI models on open-ended diagnostic reasoning, alongside a dependable framework to validate these text outputs at scale. To address this we evaluate three lightweight LLMs, Claude-Haiku-4.5, GPT-5.4-Mini, and Gemini-3.1-Flash-Lite, on free-text 5G domain knowledge and fault-analysis tasks across three benchmarks, TeleQNA ORAN FT, 5G-Faults FT, and TeleInter FT. Three independent frontier judges score outputs, and pairwise inter-judge agreement is measured as an empirical test of the LLM-as-Judge methodology. All three models reach at least 90% accuracy on fault diagnosis, while zero-shot recall of 3GPP and O-RAN specifications remains the critical gap, with all models scoring below 60%. Mean inter-judge agreement is at least 0.90 across all runs, indicating that multi-judge LLM scoring produces consistent, reproducible grades for open-ended telecom responses. Operationally, Gemini-3.1-Flash-Lite offers the best efficiency trade-off, combining competitive accuracy with the lowest inference cost and latency, making it the most suitable candidate for production telecom deployments.
Deploying high-accuracy neural networks on resource-constrained edge devices remains challenging, as existing approaches treat training, compression, and hardware synthesis as separate stages, leaving a gap between software-trained models and efficient end-to-end deployment with limited support for interpretability. We propose Bern2Edge, an end-to-end framework that uses knowledge distillation to convert a pretrained teacher feed-forward network into hardware-efficient representations via Bernstein polynomial activations. This representation enables two deployment paths: (i) a high-fidelity LUT-based realization that preserves model fidelity under compression, and (ii) a symbolic rule-based representation derived from Bernstein activation geometry, enabling interpretable inference with explicit input-space constraints. The resulting BNNs achieve up to 2.12 percentage-point (pp) accuracy improvement over ReLU under identical compression constraints. At the system level, Bern2Edge achieves up to 99.8% latency reduction and 95.2% BRAM reduction relative to a W8A8 quantized teacher on an AMD Xilinx KV260 FPGA, while maintaining accuracy within 0.5 pp, and further deploys on a low-power Spartan-7 XC7S15 FPGA. The rule-based path reduces DSP usage by up to 89.0% at a cost of 1.5 pp in total accuracy.
Non-Intrusive Load Monitoring (NILM) systems estimate individual appliance energy consumption from a single aggregate meter, without requiring separate sensors for each device. By installing a single meter that measures a building's total electricity consumption, NILM algorithms can determine the active status of each appliance. However, traditional NILM systems use computationally intensive optimization algorithms to process offline data, limiting their capability for on-device deployment, where sensitive household data must be processed locally. This paper proposes a Tsetlin Machine (TM)-based NILM framework, targeting real-time applications on resource-constrained microcontrollers (MCUs), enabling privacy-preserving edge deployment. The problem is reformulated as a classification task, and the proposed approach achieves an average precision of 90% and recall of 96% for two-appliance classification, and 77% precision and 80% recall for four appliances on the REDD dataset. The trained model occupies only 18 KB of flash memory and achieves an inference latency of 0.43 ms on an ESP32, demonstrating its suitability for embedded NILM applications on MCUs.
Nick Lemke, Ssharvien Kumar Sivakumar, Antoine P. Sanner +4cs.CV
Scene graph generation from surgical video enables a holistic and structured understanding of surgical scenes by modeling objects and their semantic relationships. Despite recent advances, state-of-the-art approaches rely on large, parameter-heavy deep learning models that are impractical for deployment in the operating room (OR) due to hardware footprint, hygiene constraints, latency, and data privacy concerns. To the best of our knowledge, this is the first scene graph generation method built on NCAs and the first NCA framework capable of learning structured representations. We introduce SG-NCA, a lightweight scene graph generation framework based on Neural Cellular Automata (NCA), designed for inference in fanless devices critical for OR hygiene protocols. SG-NCA is the first scene graph generation combining NCA-based multiclass segmentation for efficient object detection and feature extraction with a lightweight relation predictor. We evaluate SG-NCA on videos of cataract surgery and cholecystectomy, demonstrating performance comparable to established baselines while requiring 55x fewer parameters. We showcase deployment on fanless edge devices better suited for the OR and demonstrate downstream applications such as surgical video captioning, highlighting SG-NCA's potential for affordable, privacy-preserving, and OR-ready intraoperative scene understanding.
Running large AI models on resource-constrained edge devices requires model compression to reduce model size and computation. What compresses well, however, need not deploy well. We survey dozens of recent works that report compression results on real hardware and extract practical deployment guidelines from them. Following these guidelines, we deploy compact language and image models on GPU, CPU, and Raspberry Pi platforms across question answering and image segmentation. No single technique wins across tasks. For question answering, Qwen3.5 0.8B reaches 93.85 SQuAD F1 and 92 EM under Q5_K_M GGUF quantization, while structured pruning at the same precision costs 16 F1 at a 1% ratio. For segmentation, the ranking reverses: default quantization leaves parameters and MACs unchanged, whereas pruning cuts model size by nearly 80% at near-constant mIoU. Pruning can even inflate the deployed artifact by 21-49% by breaking k-quant super-block alignment; combined with longer, less format-compliant outputs, this raises Raspberry Pi latency up to 3.4x. Compression can also manufacture the appearance of competence rather than destroy it visibly: one LoRA-recovered variant stays fully parseable and holds 71% strict BoolQ accuracy while sending 97 of 100 predictions to a single class, at 52.6% balanced accuracy. We explain these effects through neural-flow graph analysis and prefill-decode-level latency decomposition, and condense them into task-specific deployment research directions. The right technique depends on the task, the model, and the hardware. Our experiment code and artifacts are open-sourced at https://github.com/Arnavvvkumar/deployment
Paul Julius Kühn, Duc Anh Nguyen, Saptarshi Neil Sinha +2cs.CV
Real-time 6DoF object pose estimation on resource-constrained hardware remains challenging, as accurate correspondence-based and refinement pipelines typically rely on non-differentiable PnP/RANSAC stages or costly iterative refinement, while recent foundation-model-based approaches incur inference costs that are prohibitive for edge deployment. We present TinyDETR-Pose, a lightweight, end-to-end, single-stage framework that jointly detects objects and regresses their full 6D pose in a single forward pass. Built on the efficient LW-DETR architecture, TinyDETR-Pose formulates detection and pose estimation as a set-prediction problem and attaches dedicated MLP heads for rotation, monocular depth, and projected object center regression to each decoder query, eliminating the need for PnP, NMS (non-maximum suppression), or iterative pose refinement. Object symmetries are handled through a ADD-S loss applied uniformly to all objects, without the need for object-specific loss schedules or separate geodesic/ADD supervision. In addition, predictions are assigned to ground truth using a symmetry-safe Hungarian matcher based on class and 2D spatial cues, yielding stable assignment under symmetry and depth ambiguity. On YCB-V, TinyDETR-Pose achieves a comparable ADD-S AUC of 85.9, while requiring up to 72.7% fewer parameters than other DETR-based single-stage pose-estimation approaches. Due to its compact design, TinyDETR-Pose runs in real time and achieves an inference latency of only ~4.5 ms per frame on an NVIDIA Jetson Nano using TensorRT, demonstrating that accurate end-to-end transformer-based 6D pose estimation can be made practical for edge deployment.
Junseo Kim, Uraz Odyurt, Amirreza Yousefzadehcs.CV cs.LG
Vision Transformers (ViTs) demonstrate exceptional performance in computer vision but suffer from large parameter counts and quadratic computational complexity, severely limiting their deployment on resource-constrained edge hardware. While recursive weight-sharing reduces parameter counts and token merging mitigates computational and memory bottlenecks, integrating these two paradigms without costly retraining is non-trivial, leaving this intersection largely unexplored. We propose MergeOver, a post-training approach that integrates Token Merging (ToMe) into the recursively weight-shared Sliced Recursive Transformer (SReT). Through an Unmerge tracking stack, constraint-safe merge-rate adjustment, and synchronised token-mass tracking across spatial permutations, MergeOver resolves the spatial and merging constraints of this integration. We further employ a stage-wise single-shot schedule that performs token reduction at the first block of each stage and maintains a fixed sequence length throughout its subsequent recursive iterations. Benchmarked on ImageNet-1K, our selected configuration reduces top-1 accuracy by 1.47 percentage points. On the GPU, it reduces peak activation memory by 37.3% and 38.4% at batch sizes 1 and 16, while throughput decreases by 21.7% at batch size 1 but increases by 21.7% at batch size 16. On a Raspberry Pi 5 (ARM CPU), it reduces latency by 2.4% and 17.6% at batch sizes 1 and 16. These results show that MergeOver can recover a meaningful part of the throughput and memory cost that recursive weight-sharing introduces, without retraining, and provides a baseline for combining token merging with hierarchical recursive transformers.
Fabric inspection in the garment industries of low-income economies remains largely manual, and commercial vision systems are priced beyond most small and medium mills. Because defects are sparse under controlled production, a natural response is a cascade: screen every frame with a cheap anomaly detector and invoke a full detector only on suspicious frames. We build such a cascade for four knit-fabric defect classes and deploy it end-to-end on an NVIDIA Jetson Nano with TensorRT FP16. Stage 1 is a compact convolutional autoencoder with decoder attention gates, an edge-weighted reconstruction loss, and feature-level distillation from a frozen YOLOv5n teacher; Stage 2 is YOLOv5n, invoked only on flagged frames. On a 249-image benchmark disjoint from detector training (20 defective, 229 non-defective), Stage 1 at a recall-prioritised threshold flags all 20 defective images (95% CI 0.83-1.00) at a false-positive rate of 49.3% (113/229), reducing false positives by 19.3% relative to a plain autoencoder (p=0.011). The parallel pipeline reaches 13.45 FPS against 9.86 FPS for a sequential YOLO-only loop. Our central finding comes from decomposing that 1.36x: 91% of it is attributable to overlapping JPEG decode with inference rather than to the cascade, which contributes only a 5.1% inference reduction at the measured forwarding rate p = 0.534. We further show that forwarding here is false-positive-limited rather than prevalence-limited - 85% of forwarded frames are false alarms - and quantify the 29-45% inference reduction attainable under tighter calibration. We report this as a caution for cascade speedups measured without controlling the data path, and position the system as AI-assisted triage rather than autonomous acceptance.
Automated plant species identification from citizen-science imagery is an established, demanding fine-grained recognition problem: large taxonomic label spaces, visually similar species, and long-tailed observations require real model capacity, while field use constrains memory, latency, and power. Model size is only part of the deployment cost: intermediate activations held in memory during inference and platformdependent execution behavior matter too, so compact recognition must be assessed on target hardware rather than through complexity metrics alone. We present BoltNet, an ultra-lightweight fully convolutional architecture combining a Spatial Redistribution Bottleneck and Logit PreSampling to improve the tradeoff between predictive performance and model size in high-cardinality classification, and report the AccuracyCompression Tradeoff as a complementary diagnostic. On Pl@ntNet300K, BoltNet reaches 0.682 F1-score with 341K parameters (1.37 MB), the highest F1-score among evaluated models below 2 MB and close to substantially larger convolutional backbones. Model-only measurements on a Raspberry Pi 5, Jetson Orin Nano, and Hailo-8 characterize execution across CPU, GPU, and NPU platforms, where BoltNet is the most consistently efficient model, with the best FPS/W on the GPU and NPU and second-best on the CPU. Results on AIDERv2 and CLRS provide secondary evidence of transfer across environmental image-classification tasks. Code available at: https://codeberg.org/danielrossi/BoltNet
Edge-deployed vision systems in target recognition, surveillance, autonomous vehicles, and drone domains require hierarchical inference pipelines where a detection model identifies objects of interest and downstream classifiers provide fine-grained attribute analysis. Running all models on the GPU creates a serial bottleneck that limits real-time throughput as pipeline stages grow. Modern edge SoCs pair GPUs with dedicated neural accelerators (NPUs, DLAs) capable of concurrent execution, yet deploying custom models on these accelerators remains impractical due to strict operator constraints, quantization incompatibilities, and an undocumented end-to-end pipeline. We target NVIDIA Jetson DLA cores as the representative platform. We present a five-step methodology for zero GPU fallback DLA INT8 deployment of classification backbones, comprising architecture adaptation, manual dynamic range workaround to rescue TensorRT's implicit quantization (recovering 94.0% accuracy from implicit quantization's 75%) for rapid pipeline validation before explicit quantization, quantization-aware training, ONNX graph surgery for DLA compilation, and a concurrent GPU-detection/DLA-classification inference pipeline. We document nine engineering constraints with root-cause analysis and generalizable solutions. Validation on a dual-head person attribute classifier running on DLA alongside a GPU object detector on a Jetson Orin NX demonstrates near-zero pipeline overhead (12.5 vs. 13.3~FPS detector-only at 1080p), with dual-DLA scaling at no additional cost. The methodology is backbone-agnostic and generalizes to any detection-classification edge pipeline.
Naveed Khan, Besan Al Sbeihi, Maryam Alshehhi +1cs.AI cs.NI eess.SP
Foundation models are emerging as a transformative paradigm for AI-native sixth-generation (6G) wireless networks by enabling scalable, transferable, and data-efficient intelligence across diverse communication tasks. Unlike conventional deep learning models that are trained for individual applications, wireless foundation models (WFMs) learn generalized representations from large-scale heterogeneous wireless data and can be efficiently adapted to communication, sensing, localization, and network optimization tasks with minimal task-specific supervision. Despite rapid progress, current research remains fragmented across architectures, training paradigms, and application domains, with no unified survey dedicated to the design, learning, and deployment of WFMs. This survey presents a comprehensive and unified review of wireless foundation models. We first establish the fundamental concepts of WFMs and introduce a taxonomy that organizes the field according to model architectures, pre-training paradigms, and applications. We then review representative architectures, self-supervised pre-training strategies, parameter-efficient adaptation methods, datasets, benchmarks, and evaluation methodologies, highlighting their roles in enabling transferable wireless intelligence. Furthermore, we examine emerging applications spanning physical-layer signal processing, network intelligence, and cross-layer optimization, and discuss the key challenges of data availability, generalization, interpretability, efficient edge deployment, and standardization. Finally, we outline future research directions toward scalable, trustworthy, and general-purpose wireless intelligence for AI-native 6G networks. This survey provides a comprehensive reference for researchers and practitioners developing next-generation intelligent wireless systems.
Edge platforms used for aerial observation must interpret aircraft imagery under limited memory, limited compute, and intermittent connectivity. This setting is difficult for standard RGB-only recognition models and general-purpose vision-language models, especially when calibrated thermal and LiDAR aircraft data are unavailable. We present TriCLE, an application-oriented tri-modal vision-language system for aircraft taxonomic grouping under edge constraints. From a single RGB aircraft image, TriCLE generates a structure-preserving FLIR-style thermal view and a pseudo-LiDAR depth projection, then fuses the aligned views with task instructions in a compact Qwen3-VL backbone. The model is aligned to an expert aircraft taxonomy based on propulsion, airframe family, size, design era, and configuration, so its outputs reflect engineering-relevant similarity rather than only surface appearance. We evaluate supervised fine-tuning, rotation-preserving SFT, and three policy-alignment strategies: GRPO, GSPO, and DAPO. Sequence-level GSPO gives the strongest validation performance, reaching 88.33\% validation accuracy and 0.91 weighted F1 on valid aircraft outputs. On a held-out aircraft test partition, GSPO achieves 78.00\% accuracy and 0.793 weighted F1 while preserving 94.00\% parseable output formatting. After 4-bit quantization and attention-memory optimization, the aligned 4B model fits an 8GB deployment target and processes each tri-modal triplet in 1.48 seconds. These results support TriCLE as a practical prototype for interpretable, edge-feasible aircraft grouping, while emphasizing the need for further validation on real aligned thermal and LiDAR sensor streams.
Chenghua Wang, Daliang Xu, Dongqi Cai +19cs.AI cs.RO
Physical AI policies require inference throughout their lifecycle, including model evaluation, cloud reinforcement learning rollout, edge GPU serving, and onboard deployment. Although these settings share the same checkpoint and action semantics, they often rely on separate inference programs. To unify them, we build PhyAI, a Physical AI inference engine with a single runtime that keeps architecture-specific conditioning, solver, cache, and output logic in model adapters while sharing graph execution, kernels, memory management, and parallel services. The same codebase runs vision-language-action (VLA) models and world-action models (WAMs) on single or multiple GPUs across onboard, edge, and cloud deployments. We used the adapter interface to add MiniCPM-Robot on the day of its release. PhyAI achieves 1.40x-4.65x speedups over the official implementations of pi0, pi0.5, GR00T N1.7, and MiniCPM-Robot. On Cosmos3-Nano-Policy-DROID it reduces latency from 2.46 to 1.18 s on eight H20 GPUs (CFG=2, TP=4), a 2.08x speedup. Specialized runtimes remain faster in several configurations, so our goal is one runtime with competitive latency rather than the fastest result in every case. Detailed profiles reveal why different models need different execution policies: on a Hopper-series GPU at batch size one, the pi0.5 action expert accounts for 8.8% of FLOPs but 57.2% of latency; at batch size 32 its share drops to 13.5% and throughput reaches about 100 samples/s. Cosmos3 remains generation-dominated and gains only 14.3% throughput as batch size increases from 1 to 16. We further introduce the control-time Roofline, which distinguishes inference-bound from environment-bound control; the measured pi0.5 points on four LIBERO suites are environment-bound while Cosmos3 stays inference-bound. Code and benchmarks: https://github.com/mingti-org/phyai.
Deploying vision-language models (VLMs) for safety-critical spatial reasoning on resource-constrained autonomous driving platforms requires both compact model size and reliable metric grounding. We present MoRAL (Multimodal Reasoning for Autonomous Language Models), a two-stage fine-tuning pipeline that teaches Cosmos-Reason2-2B to first read a physics-encoded Bird's Eye View (BEV) representation and then reason over it for driving decisions. The BEV image encodes LiDAR metric distance as color bands, object class as cluster morphology, and radar Doppler velocity as directional wedge overlays, externalizing spatial perception into the input image so that no learned 3D backbone is required at inference. Stage 1 fine-tunes the vision encoder on 60,000 grounding records; zero-shot baselines produce no parseable BEV outputs, confirming the vocabulary requires explicit training. Stage 2 fine-tunes the full model (52M parameters, 2.4% of total) on 57,696 chain-of-thought records generated by Cosmos-Reason2-8B as teacher, spanning eight driving question types. On 2,304 held-out nuScenes frames evaluated by Gemma 4 (31B) calibrated against human review, MoRAL wins seven of eight question types over a zero-shot 8B baseline despite using four times fewer parameters, with the largest margins on question types requiring structured multi-step physics reasoning. Emergency braking recall improves from 10.8% to 47.8%, output degeneration falls from 94.1% to 20.8%, and the full pipeline fits a consumer 8 GB GPU at 42 tok/s without quantization. These results establish a reproducible foundation for compact, physics-grounded VLM reasoning on mobile edge platforms.
Unified multimodal object tracking has achieved remarkable robustness by leveraging complementary sensor data (e.g., RGB, Thermal, Depth), yet the heavy computational burden of state-of-the-art models hinders their deployment on resource-constrained edge devices. In this work, we identify the prediction head as a critical but often overlooked efficiency bottleneck. By strategically streamlining the decoder architecture, we unlock the potential for real-time inference but simultaneously introduce a capacity gap between the lightweight student and the heavy teacher. To resolve this, we conduct a systematic analysis of 17 distillation strategies and introduce a Dual-Alignment Distillation framework. Our key insight is that effective compression requires decoupling knowledge transfer into two complementary streams: (1) Spatial Representation Alignment, which employs feature distillation to sharpen the student's spatial focus on foreground targets ("Where to track"); and (2) Semantic Distribution Alignment, which utilizes logit-based distillation to align decision boundaries and transfer discriminative dark knowledge ("What to track"). Extensive experiments across five benchmarks demonstrate that our approach significantly outperforms complex state-of-the-art methods. Notably, our distilled model achieves 91.5% MPR on RGBT234 and operates at 54 FPS on a single RTX 4090, representing a 5x speedup over the teacher model while maintaining superior accuracy.
Vision-Language Navigation for Unmanned Aerial Vehicles (UAV-VLN) requires rapid and reactive control in complex 3D environments. Recent minimalist end-to-end paradigms show great promise but typically rely on massive language models containing billions of parameters, incurring prohibitive latency for real-world edge deployment. In this paper, we challenge this parameter-heavy reliance. Comprehensive cross-scale evaluations reveal the critical insight that perception quality fundamentally outweighs language reasoning capacity. We demonstrate that a lightweight 2B model equipped with high-fidelity visual inputs completely matches the overall success rates of massive 7B baselines. However, this minimalist policy exposes a fundamental robustness flaw inherent to pure Behavior Cloning (BC). Lacking explicit negative feedback, the agent fails to internalize robust spatial constraints and exhibits alarming collision rates in out-of-distribution (OOD) scenarios. To overcome this vulnerability without relying on unscalable human annotations, we propose AeroDPO, a zero-cost automated Direct Preference Optimization pipeline driven by deterministic physical simulation state rollback. Upon detecting collisions, the system autonomously rewinds the environment to extract causal reasoning errors as rejected actions, applies decoupled privileged interventions to synthesize collision-avoidance preferred maneuvers, and leverages an offline vision language inspector to filter visual ambiguities. By equipping our 2B model with this automated data flywheel, AeroDPO boosts success rates to 49.16% on unmapped scenarios while drastically suppressing collision rates, establishing a new SOTA for autonomous aerial agents.
Cybersecurity is a fundamental requirement for protecting wearable devices used in healthcare Internet of Things (H-IoT) systems. Security failures in these resource-constrained systems directly compromise patient safety. Physiological data and network traffic are frequent targets of cyberattacks in H-IoT environments. To address these risks, deep learning-based cybersecurity mechanisms for H-IoT often involve complex architectures with large parameter counts. Existing datasets are also rarely assessed for quality, limiting their applicability. However, this research addresses these challenges by developing multiple realistic datasets and proposing lightweight deep learning models, namely the Temporal Convolutional Network (TCN) and Residual TCN (Res-TCN), for H-IoT. It includes two binary classification datasets for Distributed Denial of Service (DDoS) attacks and a multiclass dataset representing Selective Forwarding (SF), Man-in-the-Middle (MITM), and DDoS attacks. The datasets UL-ECE-MQTT-DDoS-H-IoT2025 and UL-ECE-UDP-DDoS-H-IoT2025 are generated in Cooja and ns-3 to capture transmission behaviours and protocol variations. The third dataset, UL-ECE-MultiAttack-H-IoT2025, integrates physiological and network features to represent multiple cyber threats in H-IoT. Building on this, the TCN model is designed to detect and mitigate DDoS attacks over the MQTT and UDP-based datasets. It incorporates a monitoring frequency-based detection mechanism and a dynamic threshold-based mitigation strategy. To enable edge deployment, the model is quantised and converted into TensorFlow Lite (TFLite) for real-time DDoS detection on Raspberry Pi 4, achieving low latency and power-efficient operation in H-IoT. This thesis establishes a deep learning-based cybersecurity defence mechanism encompassing realistic dataset generation, lightweight model design, and edge deployment for securing H-IoT systems.
LLM agents following the ReAct paradigm are promising enablers of complex multi-step tasks, including multi-hop question answering, code generation, and control of physical AI systems. Yet, when deployed at the edge, they must tightly manage their reasoning budget while remaining reliable and deferring to a cloud-side model only when local uncertainty is too high to act safely. We propose Think Short, Defer Smart (TSDS), a framework that synergistically integrates a lightweight convergence probe, which halts on-device reasoning once the intended action has stabilized, with a perplexity-based deferral rule that escalates uncertain actions to a cloud-side model. Both mechanisms are jointly calibrated on end-to-end episode trajectories via a multi-objective Learn-Then-Test (LTT) procedure, providing simultaneous finite-sample guarantees on expected episode reward and cloud-call rate. We evaluate TSDS on four ReAct benchmarks spanning arithmetic reasoning (GSM8K), multi-hop question answering (HotpotQA), code generation (MBPP), and multi-step embodied planning (household robot), and compare against thought-calibration-only and calibrated-deferral-only standalone baselines. TSDS reduces per-episode thinking compute by 43%-65% over deferral-only baselines across HotpotQA, MBPP, and the household robot task, while maintaining certified reward and cloud-call rate guarantees.